• Log in with Facebook Log in with Twitter Log In with Google      Sign In    
  • Create Account
  LongeCity
              Advocacy & Research for Unlimited Lifespans

Photo
- - - - -

F@H Productivity & Performance


  • Please log in to reply
26 replies to this topic

#1 dnamechanic

  • Life Member
  • 1,518 posts
  • 0
  • Location:Dallas, TX

Posted 01 April 2008 - 11:47 PM


This thread is aimed at questions that may arise regarding expected productivity for a particular computer.

As you may expect there are multiple factors determining productivity. In this posting, productivity may be referred to a points-per-day (PPD).

Some of the major factors are:

Client software - The type of folding client that was downloaded from Stanford site.

Work units - regular uniprocessor work units, bonus work units, large work units, advanced methods work units, SMP work units, special workunits for PS3s or graphics processors.

Nomenclature - Work units have a name, one can determine what work unit is being processed by looking at a client display, a work unit title might look like: p4419_Seq_50_unf_AMBER. Usually only the first four numbers (p numbers for protein) are required to classify the unit, in this case 4419. Stanford has a list of current work units in process here. One can look up 4419 and see that it is valued at 15 points, etc.

Operating system- Major folding OS'es are WindowsXP, Windows Vista, various forms of Linux, & Mac OS'es

Computer hardware - Personal computers (PCs - once known as IBM clones), Macintosh, Notebook (Laptop), Servers (usually running some form of Windows or Linux), Playstation 3 (PS3), Graphics Processors (currently ATI only)

Computer hardware can be divided into classes depending on numbers of processor cores:

- single core
- dual core
- quad core
- octal (two quad cores on one motherboard, common in servers)

Probably, the most common combination for a regular computer is a single-core processor running Windows XP (or Vista, or Win 98) folding a regular work unit. Stanford benchmarks such work units on a 2.8 GHz Pentium-4 system and the points for work units are set at 110 points-per-day (PPD). So, if your computer runs faster or slower than the benchmark machine then you can scale your expected PPD accordingly.

If your hardware differs, then your expected PPD will be different. In general, processors in the P-4 class and older are a bit slower clock-for-clock than their corresponding AMD counterparts. In the newer processors, Intel Core 2 processors are the faster folders than corresponding AMD versions. Recent Mac computers use the Core 2 or even newer versions of Intel processors and are exceptionally good folders. Older Macs of the G5 variety are pretty close to P4 processors, depending on type of work units, etc. Mac G4's are quite slow.

A web site called FahInfo collects performance data from all types and combinations and complies it into a useful database.

You can search for :

- a particular work unit (use the 'p' number described above) and see performance that various computers obtain.
- compare different processors (Intel vs AMD vs Mac)
- average scores for all work units that have been reported
- other useful data

http://www.fahinfo.org/index.php

Likely there are related productivity topics and items not in the lists above, feel free to mention them in this thread.

Some, or all, of the above productivity factors can be discussed in more detail.

Edited by dnamechanic, 02 April 2008 - 01:00 AM.


#2 dnamechanic

  • Topic Starter
  • Life Member
  • 1,518 posts
  • 0
  • Location:Dallas, TX

Posted 09 April 2008 - 06:10 PM

Bonus-Point Work Units - Up to twice PPD

As alluded to in previous performance summary, all work units are not equal. For the typical uniprocessor computer, one of the best ways to increase productivity (points per day) is to fold bonus point WUs. As mentioned before, Stanford benchmarks regular work units at 110 PPD. But, certain WUs are credited with as much as twice regular points.

Because the bonus-point WUs place a greater demand on the computer and/or the Internet bandwidth, a contributor has to request them before they are sent. Probably the best bonus-point WUs for a typical contributor to request are the WUs known as "BigWUs". BigWUs as you might expect are bigger, usually the download file size is larger (maybe 3-4 MB), or the running memory (RAM) requirements are larger (up to maybe 150 MB RAM). Of course, the resulting file to send back to Stanford is typically large, usually 7-9 MB (although have seen as large as 40 MB).

The good thing is that the contributor gets additional credit for these additional demands, probably around 2x (twice) normal PPD.

So, if you have a broadband connection (essentially non-dial-up) and extra system memory, these 'BigWUs' are good to increase PPD. The extra memory requirement can be quite modest for Windows XP or even Win98 (512 MB total usually works fine, unless you work with memory intensive applications or regularily have many applications open simultaneously). Systems will work most of these BigWUs with only 256 MB RAM, but requires special attention, and no other applications running. So, running these larger work units with 256 MB is *not recommended*. For Windows Vista, RAM requirements in general are increased, most likely any system configured for Vista will have sufficient RAM.

To receive these "BigWUs" - You must set the client configuration to accept them. This is usually clicking 'Yes' to accept Big Work Units or Accept WUs>5MB in your configuration setup.

Having your client configured to receive bonus-point WUs does not guarantee that you will receive them. Often, they are simply not available at the time your client may request a WU. In some clients, particular WUs are assigned according to your computers capability to process them in a timely manner (as determined via CPU ID, RAM size alloted, or even a quick speed test).

Edited by dnamechanic, 09 April 2008 - 06:49 PM.


#3 dnamechanic

  • Topic Starter
  • Life Member
  • 1,518 posts
  • 0
  • Location:Dallas, TX

Posted 12 April 2008 - 05:21 PM

Monitoring F@H Performance

To maximize productivity, it is often helpful to be able to monitor the ongoing folding process. Stanford provides rudimentary monitoring in the downloaded client software. But, a third-party monitoring utility can provide more, such as:

- Estimated time of work unit completion (ETA).
- Points Value of Work Unit
- Points-per-Day (PPD)
- Core Type- Can often indicate whether bonus point WU or not
- Deadlines – both Preferred and Final
- Monitor multiple clients (for multiple computers or CPUs, either locally or over network)
- Quick info at a glance (progress, points per WU, etc)
- An alert, if client gets stalled or hung-up (can help avoid lost folding time)

A few dedicated folders have developed monitor utlilties for the F@H clients. They all have good and often unique features. Currently, the most popular monitoring utility probably is FahMon.

FahMon automatically displays all parameters listed. It even recovers your username and team number from actual folding work files, so that you can be sure that your work is credited correctly.

This is a screenshot of FahMon when monitoring a single client:

FahMon_Demo_1.jpg

Below, FahMon is monitoring two clients on two different computers, via network.

The second client is selected (blue) and the pertinent info for that client is showing in the text at the right. The total PPD for both clients is displayed at the end of the progress bar at the bottom:
FahMon_Demo_2.jpg

The FahMon site has good instructions and screenshots to help you set it up.

You may download FahMon here:

http://fahmon.net/download.html

Edited by dnamechanic, 12 April 2008 - 06:52 PM.


#4 Live Forever

  • Guest Recorder
  • 7,475 posts
  • 9
  • Location:Atlanta, GA USA

Posted 16 April 2008 - 07:39 PM

Anyone wanting to try to step it up some might try out the new GPU2 out just 2 days now: http://folding.typep...tes-to-the.html

Since it is out, those running GPUs are about to surpass the entire contributions of all "regular" Windows users, with less than 1% of the amount of active contributors: http://fah-web.stanf...y?qtype=osstats

#5 Mariusz

  • Guest
  • 164 posts
  • 0
  • Location:Hartford, CT

Posted 18 April 2008 - 05:47 PM

I 've run folding@home client on 4 computers, and still I'm not able to achive results as some people with one PS3.
In addition to that, there is more important issue - the cost of folding.
Well it seems that its costing me about $12 per each computer per month!
Sorry guys but I think that $50/month could be spend better. (SENS for example)

And with 62 Longevity Meme Folding@Home members in competition we are burning a lot of cash for something could be done with higher efficiency by one or two supercomputers.

Would you like to know more?:D
"it does cost you a little cash out of your pocket each month to pay for the electricity. How much, you ask? That's what we're here for."
http://ps3.ign.com/a...6/776347p1.html


System power consumption
Idle with Cool'n'Quiet 98.6W
Idle without Cool'n'Quiet 113.0W
CPU client x1 160.2W
CPU client x 2185.6W
GPU client 195.6W
CPU client + GPU client 228W
http://techreport.co...icles.x/11022/1


http://www.xyhd.tv/2...ing-the-planet/
Play Station 3 is doing a lot of folding in its spare time, but at 200 watts of draw it is consuming just over 6 Kilowatts a day of power. This is roughly 10 pounds of Coal I rounded down to make the math easy. When coal is burned it combines with Oxygen to form the Green House Gas, Carbon Dioxide. Each Pound of Coal burned adds 2 pounds of Carbon Dioxide to the air, and removes 1 pound of Oxygen. The cost to the planet of using 3650 pounds of Coal per year times 200,000 folders… is huge, but if you just want a monetary number that is $12,775,000 worth of coal at $35 a ton, or based on the price of 12 cents a kilowatt it is $70,080,000 and 730 kilotons of Carbon dioxide.

Edited by Mariusz, 18 April 2008 - 05:55 PM.


#6 Live Forever

  • Guest Recorder
  • 7,475 posts
  • 9
  • Location:Atlanta, GA USA

Posted 18 April 2008 - 06:58 PM

I 've run folding@home client on 4 computers, and still I'm not able to achive results as some people with one PS3.
In addition to that, there is more important issue - the cost of folding.
Well it seems that its costing me about $12 per each computer per month!
Sorry guys but I think that $50/month could be spend better. (SENS for example)

And with 62 Longevity Meme Folding@Home members in competition we are burning a lot of cash for something could be done with higher efficiency by one or two supercomputers.

Would you like to know more?:D
"it does cost you a little cash out of your pocket each month to pay for the electricity. How much, you ask? That's what we're here for."
http://ps3.ign.com/a...6/776347p1.html


System power consumption
Idle with Cool'n'Quiet 98.6W
Idle without Cool'n'Quiet 113.0W
CPU client x1 160.2W
CPU client x 2185.6W
GPU client 195.6W
CPU client + GPU client 228W
http://techreport.co...icles.x/11022/1


http://www.xyhd.tv/2...ing-the-planet/
Play Station 3 is doing a lot of folding in its spare time, but at 200 watts of draw it is consuming just over 6 Kilowatts a day of power. This is roughly 10 pounds of Coal I rounded down to make the math easy. When coal is burned it combines with Oxygen to form the Green House Gas, Carbon Dioxide. Each Pound of Coal burned adds 2 pounds of Carbon Dioxide to the air, and removes 1 pound of Oxygen. The cost to the planet of using 3650 pounds of Coal per year times 200,000 folders… is huge, but if you just want a monetary number that is $12,775,000 worth of coal at $35 a ton, or based on the price of 12 cents a kilowatt it is $70,080,000 and 730 kilotons of Carbon dioxide.



You should use the GPU client if you want to outperform PS3s. http://folding.typep...tes-to-the.html

The new GPU client is now out, and the average person running it way outperforms (about 2 to 1) PS3s.

Edited by Live Forever, 18 April 2008 - 07:00 PM.


#7 Mariusz

  • Guest
  • 164 posts
  • 0
  • Location:Hartford, CT

Posted 18 April 2008 - 07:04 PM

The new GPU client is now out, and the average person running it way outperforms (about 2 to 1) PS3s.


It still will cost about the same, and have same result on longevity research - none.
And why was my topic moved?

Mariusz

#8 Liquidus

  • Guest
  • 446 posts
  • 2
  • Location:Earth

Posted 18 April 2008 - 07:06 PM

I've always been under the impression that f@h was intended to be used when users are not using their computers. It's great that people dedicate computers and raw processing power to initiatives like F@H, but I don't think they ever really expected people to do that. Having the program run occasionally as a 'screen saver' is what most people do, and in that case, the costs aren't so much.

I'm not going to advocate illegal computer use, but there are so many servers world wide that host irrelevant data that hardly gets touched, if all that unused processing power was harnessed, you could really put a dent into data collection.

#9 Live Forever

  • Guest Recorder
  • 7,475 posts
  • 9
  • Location:Atlanta, GA USA

Posted 18 April 2008 - 07:23 PM

The new GPU client is now out, and the average person running it way outperforms (about 2 to 1) PS3s.


It still will cost about the same, and have same result on longevity research - none.
And why was my topic moved?

Mariusz

Well, if you would rather spend your computer cycles, money, etc on something else, then you are free to. It is just a nice thing to run on your computer when you aren't using it to maybe help out a little. I PM'd you about your topic being merged.

I've always been under the impression that f@h was intended to be used when users are not using their computers. It's great that people dedicate computers and raw processing power to initiatives like F@H, but I don't think they ever really expected people to do that. Having the program run occasionally as a 'screen saver' is what most people do, and in that case, the costs aren't so much.

I'm not going to advocate illegal computer use, but there are so many servers world wide that host irrelevant data that hardly gets touched, if all that unused processing power was harnessed, you could really put a dent into data collection.

Yeah, that was the impression I got as well. It is just a nice little something to run in the background while you have your computer (or PS3) on but aren't using it. (since it would be consuming power anyway from being on) Some of these guys running server farms or multiple PS3s dedicated solely to folding are great, don't get me wrong, but the vast majority of people are just taking a "it is better than not running anything on my computer (or PS3)" approach, I think.

No one is forcing anyone to participate, though. If someone feels like their resources can be spent better doing something else, then they are more than welcome to do that thing instead. :)

#10 dnamechanic

  • Topic Starter
  • Life Member
  • 1,518 posts
  • 0
  • Location:Dallas, TX

Posted 18 April 2008 - 10:01 PM

Mariusz,

You bring up some interesting issues:

- Performance of regular computers relative to PS3s

- Dollar cost of electricity for folding

- Efficiency of various folding platforms

- Folding cost in terms of atmospheric carbon

- Effect of folding on longevity research

These issues are diverse, but all are valid topics related to folding. Live Forever and GSnake also added good information to these subjects. Perhaps exploring them one at a time might be a good approach. I will not claim to be able conclusively address every one of these, or any of them for that matter. Others may wish to weigh in on the subjects.

In any event, you have made considerable contribution to the team, to Stanford, to scientific research, and to the Meme, and that is appreciated.

#11 naapi

  • Guest, F@H
  • 49 posts
  • 0
  • Location:Poland

Posted 19 April 2008 - 01:02 PM

Mariusz,

Thanks a lot for your post, I think it touches on many points that many of us have considered for themselves.

I, too, had a lot of doubts as to whether I should contribute to SENS program or to the Longevity Meme Team. Therefore I decided contribute to both. By now I have spent somewhat more on the FAH program than on SENS, but I have committed much more to SENS and will mot likely spend much more on it within the next 2-5 years.

There are several reasons why I decided to spend money on TLM: 1) MFoundation is relatively well financed with 11 million dollars donated or commited, 2) I consider TLM a great way to indirectly promote longevity science and SENS, 3) I believe FAH has tremendous potential to become a major longevity science driver in the intermediate future (10-20 years time) 4) I can make much more difference with my couple of thousands dollars for FAH/TLM rather than for SENS.

Let me touch on all of them one by one.

1) As you likely know MFoundation has now over 11 million dollars donated or commited. The increase of funds has recently been steady and with the commitments of people belonging to the 300 group, the stream of funds is secured for the years to come. True, much more is needed - likely billions before we will have cures for aging-related diseases - but for the next couple of years SENS seems financed relatively well.

2) Lots of folders look at the stats and likely get interested in teams that occupy top positions and grow fast. As these are people interested in fighting diseases, they will likely get interested in fighting aging as well.
TLM prize is a great idea. It makes our team very special - I do not think there are many (if any) teams that offer cash prizes for folding. If anyone is to join FAH with good computing power, our team is preferred to many others. Getting to top 100 is important - it is the first page of Extremeoverclocking.com stats and, you know, being in top 100 teams out of 120 thousands or so sounds good - these nice-round numers...
Based on Kakaostats.com we will reach 150th position by yearend. This will likely happen faster, if we increase our computer power - hopefully at speed comparable to the past three weeks. If we get to 100 mark lots of people will ask a question: what kind of meme is that, that is stronger than even some of the national memes (like Canada). They might get interested and find out about MFoundation, SENS etc. All this will be accomplished with advertising dollars spent for a worthy cause rather than commercial ads.

3) This is a topic that I thought a lot about over the past two weeks or so. We now simulate proteins assembled from roughly 100 thousand atoms. Question: What would it take to simulate a whole living cell? I did a quick math and, following some assumptions concerning Moore's law, cell sizes, increase in number of folders and certain model optimizations, I came to a conclusion that simulating a whole cell will likely be possible within the next 15-20 years.

Just think about this.

We could introduce any molecule into such model and check what are the implication for metabolic pathways. We could try treatments in-silico before we would try them in-vitro or in-vivo. I consider the legal problems and risk relating to people dying during experimental therapies as one of the biggest risks of the SENS. If we could simulate what we want to do in real life, that would likely silence the skeptics.

And, while 15-20 years time frame might seem a long time - it is not. We will likely reach Robust Mice Rejuvenation in about 10 years, only afterwards committing the SENS to human research.

I can post the calculations, if anyone's interested.

4) Three previous reasons were all valid and rational ones - in my opinion. This one is a bit emotional and personal. I do not have a million dollars to spare and donate to SENS to make a 10% difference. But I can spend a thousand dollars and make a 10% difference to TLM team. With 10 thousand dollars I could be one of the biggest folders in the world and take our team to top 50 teams of the world. It is less motivating to be responsible for 0.02% of a project than for 5% of another one. Anybody needs at least some motivation and even if a cause is proper, one needs to feel that they are meaningfully contributing. Everyone needs to feel special, not to be one of the many.

As to your comments concerning the use of electricity - it is all true, that folding on dedicated computers uses fossil fuels and contributes to global warming. Following the discussion by Bjorn Lomborg on TED, I am much less concerned about couple of islanders (possibly) relocating from Tuvalu or the Maldives 100 years from now. I am concerned about 100 thousand people dying each day now, as I'm sure all of us here do.

I would like to recommend to everyone two presentations from TED. They are eye opening. If you have not seen them, these will likely be one of the best spent 40 minutes of your life:

http://www.ted.com/i...alks/view/id/44
http://www.ted.com/i...alks/view/id/62

Thanks for listening,

Maciek.

Edited by naapi, 19 April 2008 - 01:04 PM.


#12 Mind

  • Life Member, Director, Moderator, Treasurer
  • 18,997 posts
  • 2,000
  • Location:Wausau, WI

Posted 19 April 2008 - 02:30 PM

On the question of carbon footprint, everything everyone does leaves a footprint (first and foremost because humans exhale CO2). Because we are living entities we use resources. So it comes down to how you use resources that matters. I would rather use up some electricity running Folding@home than many other activities because it will benefit human beings (and the planet) in the long run. Almost every research endeavor will help, whether it be Folding@home, nanotech research, solar power research, etc... I would rather engineer solutions to our problems than restrict progress (as it seems the global warming crowd is want to do).

SO FOLD AWAY!

#13 dnamechanic

  • Topic Starter
  • Life Member
  • 1,518 posts
  • 0
  • Location:Dallas, TX

Posted 20 April 2008 - 01:31 PM

Welcome to ImmInst, Naapi (Maciek).

It was inspiring to read your post. It is clear that you put a lot of thought into your message.

I can appreciate your logic and your rationale for taking action.

Thank you for your participation in the folding effort with the F@H Prize and for signing on as a Three Hundred Member with the Methuselah Foundation.

----------------------------------------------------------------------------------------

Mind, good insight about the production of carbon dioxide. I agree with your premise that technology is the solution. It always has been.

#14 dnamechanic

  • Topic Starter
  • Life Member
  • 1,518 posts
  • 0
  • Location:Dallas, TX

Posted 21 April 2008 - 11:23 AM

Performance & Efficiency - PS3s and Regular Computers

PS3s have a different architecture than regular computers. PS3s use a processor more attuned to parallel processing related to streaming processing. Regular computers use a general-purpose processor (CPU) that can essentially compute anything, but are computationally inefficient at almost everything relative to specialized processors. Currently, regular general purpose CPUs operate at a high speed relative to main memory speed and rely on fairly large and fast hardware cache memory as a buffer to help keep data supplied to the CPU. Graphics processors, of which the PS3 is related, have little or no hardware cache and the processor calculations must be closely matched to memory (RAM) throughput. This latter requirement can sometimes be a complete showstopper.

As a result of these and other considerable differences, PS3s and graphic processors (GPUs) cannot process all types of work units. The work units they can process, they can do very fast. The PS3 in terms of FLOPS far exceeds what is available from the fastest regular computers. If the PS3 were awarded points for working work units based on FLOPS alone, then they would be awarded considerably more points than they are now (~1,000 PPD). But alas, the PS3 can do only specialized work units and the F@H program needs all kinds of work units processed. Stanford cannot afford to discourage contributions from regular computers.

Folding contributors may see 1,000 PPD from PS3s and compare to 110 PPD from a somewhat typical single-core computer and then be discouraged. But regular computers can process all types of work units and the PS3 cannot do this. The regular computer is still very important and is needed.

Many work units for regular computers offer bonus points. Depending on the CPU in a regular computer and the work unit type, up to 1,000 points a day can be earned with a regular computer. For example, the second figure in a post above (about monitoring F@H) shows a computer folding a p2615 work unit at the rate of 565 PPD. This work unit is being processed on an approximately 3-year-old notebook computer with a single-core Pentium M CPU (2.1 GHz).

A computer with a dual or quad core CPU running multiple regular clients can considerably increase PPD compared to running a single client. The best utilization of multiple-core CPUs is folding SMP work units. To fold SMP work units requires downloading a different core and currently requires a bit of extra care in the installation, it is still hasn't had all the bugs worked out (but it is pretty clean). Running the lowest-cost Intel Core-2 quad processor (Q6600) at stock speed of 2.4 GHz and with a set-up aimed for optimum efficiency yields ~3,500 PPD while consuming ~110 Watts for the total system. A more typical setup will at least yield 2,200 PPD for similar power dissipation. Newer, or more expensive, Intel quad processors will produce more PPD. Additionally, many contributors overclock their processors for an approximately 50% increase above baseline numbers, although power efficiency may degrade with overclocking. Other processors, namely dual-core CPUs, can be very effective. Currently AMD CPUs are computationally less efficient than Intel's when processing SMP work units, but a modest dual-core AMD 5200+ system can yield 1,100 PPD. This exceeds the PPD of a PS3 and depending on the system, power consumption can be considerably less than a PS3. The original PS3 consumes about 200 Watts while folding. A recent redesigned version of the PS3 with a 65 nm processor uses less power, approximately 120 Watts. A modest, although probably not typical, AMD dual-core system can fold 1,300 PPD while consuming 75 Watts.

Summary of PS3 & Regular Computer (PPD & Power consumption):

Processor........................ ~PPD..... ~Watts.....Comments
PS3 (90 nm processor)...... 1,000........200.........reported, see notes
PS3 (65 nm processor)...... 1,000........125.........reported, see notes
Intel Q6600 Quad.............. 3,500.......110........efficiency optimization
Intel Q6600 Quad............ >2,200.......125........approximately typical setup
AMD 5200+ Dual-core......1,100........120 .......approximately typical
AMD 5200+ Dual-core......1,300.........75.........semi-optimized

Notes:
- PS3 figures from posts elsewhere, here and here.
- Intel & AMD figures measured while folding SMP work units.

Edited by dnamechanic, 21 April 2008 - 11:12 PM.


#15 kismet

  • Guest
  • 2,984 posts
  • 424
  • Location:Austria, Vienna

Posted 19 August 2008 - 12:38 PM

In the last ~24hours our PPD dropped by 10'000 http://www.imminst.o...o...0&start=120
Could it be due to those networking errors? I get them regularly, the gpu client can't connect to a server and idles. Is this global issue? Or is it just my networking (when I turn off the modem the same happens)? 

Even worse the smp client errors out and I have to reinstall everything from scratch due to "Missing_Work_File" errors. I know it's beta crap, and a beta is never bug-free, but damn it, the high-performance clients contribute the most points to F@h, can't they fix it faster? F@h is buggy and everything but userfriendly, nonetheless it's close to 3 PetaFLOPS, imagine what happens when all the bugs are ironed out..

#16 sentinel

  • Guest, F@H
  • 794 posts
  • 11
  • Location:London (ish)

Posted 19 August 2008 - 01:12 PM

In the last ~24hours our PPD dropped by 10'000 http://www.imminst.o...o...0&start=120
Could it be due to those networking errors? I get them regularly, the gpu client can't connect to a server and idles. Is this global issue? Or is it just my networking (when I turn off the modem the same happens)? 

Even worse the smp client errors out and I have to reinstall everything from scratch due to "Missing_Work_File" errors. I know it's beta crap, and a beta is never bug-free, but damn it, the high-performance clients contribute the most points to F@h, can't they fix it faster? F@h is buggy and everything but userfriendly, nonetheless it's close to 3 PetaFLOPS, imagine what happens when all the bugs are ironed out..


Not sure where you are Kismet but I'm in London and probably picked the wrong evening (last night) to try my first GPU install. Just kept "waiting for work". In the end i got fed up and re-installed the CPU client before goin gto bed so as not to lose 8+ hous folding. Still don't know if it's the network or my drivers which is frustrating. I seem to have at least 1 WU "owing" as my psoition has remained static but I gues I have to be big about it and remind myself that the research comes first. But as you would no doubt agree - less bugs = better performance = more research.

#17 dnamechanic

  • Topic Starter
  • Life Member
  • 1,518 posts
  • 0
  • Location:Dallas, TX

Posted 19 August 2008 - 01:22 PM

In the last ~24hours our PPD dropped by 10'000 http://www.imminst.o...o...0&start=120
Could it be due to those networking errors? I get them regularly, the gpu client can't connect to a server and idles. Is this global issue? Or is it just my networking (when I turn off the modem the same happens)? 


Good observation, kismet.

This happens.

Stanford has a fairly large network of servers. Some servers are dedicated to sending work units and others dedicated to receiving/collecting completed WUs, and some servers do both. Then there is the consolidation of data (WU scores). If there is a glitch, sometimes reported/completed WUs are delayed. Probably the points glitch/delay will correct itself in a few days.

I have recently seen some new SMP work units that are quite difficult to fold and therefore the PPD has fallen for those work units. Don't yet know how prevalent they are.

By the way, some time ago I read that many of their servers and related equipment were donated to them.

Even worse the smp client errors out and I have to reinstall everything from scratch due to "Missing_Work_File" errors. I know it's beta crap, and a beta is never bug-free, but damn it, the high-performance clients contribute the most points to F@h, can't they fix it faster?


The Pande group at Stanford are chemistry/biology specialists. That is their main objective. A good amount of the support is completely voluntary. And much of the work is done by already overworked graduate students, trying to make a difference while getting a Masters/PhD degree. Much of the computer-related support is by the graduate students themselves.

One of the most criticized of their projects thus far has been the Windows SMP client (still in beta). The Windows OS structure is very unfriendly to the SMP processes needed for folding. And, to my knowledge, they are getting no help from Microsoft. The Linux SMP client is very stable, and folds faster than the Windows version.

F@h is buggy and everything but userfriendly, ..., imagine what happens when all the bugs are ironed out..

They do not stand to make millions off this project so they cannot budget a team of software engineers to make the code really slick. Again this is a relatively small non-profit operation. Actually I think the regular client and the Linux client are pretty easy stuff to deal with. I have a relative that has almost no computer skills, (has problems with email). He folds, with almost no help from me, and seemingly gets a feeling of accomplishment from it. This is possible because the regular client is pretty bug-free.

... nonetheless it's close to 3 PetaFLOPS...


It is a great accomplishment. They are most surely working on a relative small budget.

#18 kismet

  • Guest
  • 2,984 posts
  • 424
  • Location:Austria, Vienna

Posted 19 August 2008 - 01:47 PM

 
The Pande group at Stanford are chemistry/biology specialists. That is their main objective. A good amount of the support is completely voluntary. And much of the work is done by already overworked graduate students, trying to make a difference while getting a Masters/PhD degree. Much of the computer-related support is by the graduate students themselves.

It may have sounded cynical, but I do mean it in a positive way. It's such a damn nice project, it's a great proof of concept and it can get even better, there are so many computers that idle, they could easily obliterate the Top500 performance-wise. Why can't they get more funds, more money to offer better support...

One of the most criticized of their projects thus far has been the Windows SMP client (still in beta). The Windows OS structure is very unfriendly to the SMP processes needed for folding. And, to my knowledge, they are getting no help from Microsoft. The Linux SMP client is very stable, and folds faster than the Windows version.

Is the 64bit/vista version any better? I guess not.

#19 dnamechanic

  • Topic Starter
  • Life Member
  • 1,518 posts
  • 0
  • Location:Dallas, TX

Posted 19 August 2008 - 02:23 PM

It may have sounded cynical, but I do mean it in a positive way.

Having seen your previous interest in the subject cleared away my perception of any potential cynicism.

It's such a damn nice project, it's a great proof of concept and it can get even better, there are so many computers that idle, they could easily obliterate the Top500 performance-wise.

Yes, home computers represents a great untapped resource throughout the world.

Why can't they get more funds, more money to offer better support...


Probably for the same reasons that ImmInst, Methuselah Foundation, SENS, .. on and on, cannot get more support.

Its partially related to the psychologically studied "herd behavior". We (human beings) often have the illusion that we are independent thinkers, but the truth is the vast majority follow the herd. Until something becomes so well known that it is almost common knowledge among the populace, it gets ignored by all but the true innovators, which are very very few indeed.

Here at the ImmInst, many are curious but are unwilling to reach out (unwilling to do things directly related to what is apparently their own goals and that would benefit humankind, even at almost no cost).

For a current example, see the threads on MFURI.

Is the 64bit/vista version any better? I guess not.

Apparently not. Vista has the same problem as WinXP regarding teh SMP client. The difficulty resides in how multiple CPUs/cores are allowed to communicate with each other. Its related to the MPI (see Wiki) in the MPIEXEC.EXE module in the SMP application folder.

#20 sentinel

  • Guest, F@H
  • 794 posts
  • 11
  • Location:London (ish)

Posted 19 August 2008 - 04:42 PM

Don't know if you were aware of this already but this would explain som eof the problems we seem to have had re connections and Wu updates:

From the Stanford folding forum:
Stanford Power Outage
on Mon Aug 18, 2008 7:42 pm

"Stanford just got hit by a campus-wide power outage. We don't have any further information or an ETA at this moment, but a number of our servers are down. assign.stanford.edu appears to be up on emergency power at this time."

They went on to say that power was back a couple of hours later but that they were not going to try and restart several until the server-rooms had cooled down.

Edited by sentinel, 19 August 2008 - 04:43 PM.


#21 sentinel

  • Guest, F@H
  • 794 posts
  • 11
  • Location:London (ish)

Posted 10 September 2008 - 02:07 PM

Hi Chaps

I've been leafing thru but couldn't find it. Could anyone say how you can retrospectively configure for Big WUs or do you have to wait until you have finished a unit then re-install and insert Y when it is provided as an option?

I've retro'd the -admethods via the shortcut methods and just wanted to see if the same was possible for Big WUs.

Cheers

Sentinel

#22 sentinel

  • Guest, F@H
  • 794 posts
  • 11
  • Location:London (ish)

Posted 10 September 2008 - 03:04 PM

Hi Chaps

I've been leafing thru but couldn't find it. Could anyone say how you can retrospectively configure for Big WUs or do you have to wait until you have finished a unit then re-install and insert Y when it is provided as an option?

I've retro'd the -admethods via the shortcut methods and just wanted to see if the same was possible for Big WUs.

Cheers

Sentinel


Ignore me I'm a moronic ludite (look it up). I've sorted the reall quite obvious, tick-box answer myself. I also appear to talk to myself. "Oh, look, a reply to my topic! Oh yes, It's the one I typed 2 minutes ago..."
Sigh.

Edited by sentinel, 10 September 2008 - 03:14 PM.


#23 dnamechanic

  • Topic Starter
  • Life Member
  • 1,518 posts
  • 0
  • Location:Dallas, TX

Posted 04 January 2009 - 05:11 AM

This question was initiated in this thread.

Thanks fizzionz, for your efforts and welcome to the TLM team.

ok, got it running mint. but i am wondering. will my quad core @ 2.5ghz crunch more data then if i use the GPU client with my asus GeForce 7600 GS? im pretty sure the GPU has 512 mb ram.

Most folders with a quad core CPU find that a combination of GPU and SMP folding provide the most PPD.

This is particularly so when running the Windows SMP client (since it is less efficient than the Linux client).

allthough the quad core has access to the max that xp will allow. i have 4 gb ram but only 3.something is used. and the folding @home doesnt use anywhere near that. im running at 93% cpu speed with like 840mb ram used.

Yes, a single SMP client can use anywhere from 200MB to 700MB RAM, although the performance doesn't seem to suffer when RAM is limited to less than the client would use. In other words, if excess RAM is available a client can capture more than 700MB, but if less is available (that is to some lower limit then it will fold with about the same efficiency).

so yeah. will the gpu preform better then the quad core?

To run the GPU client, the Graphics card must be capable of running CUDA.

Your 7600GS is not listed as CUDA capable (List here).

If you want to stay with Windows, you might consider running two SMP clients with an Affinity Changer to raise the computational efficiency. (mention of another Affinity changer.

Intel Quad processors, until the I7 version, were essentially two dual-cores together on the same IC. The idea of changing affinity is to assign the threads of each of two SMP clients to the appropriate two cores of the quad (ie to the cores associated together). This helps improve points-per-day for the Windows SMP client.

Edited by dnamechanic, 04 January 2009 - 05:17 AM.


#24 fizzionz

  • Guest, F@H
  • 32 posts
  • 0

Posted 10 January 2009 - 04:08 AM

ok so, ive got t he quad running. but i install the affinity changer but cannot run two instances of the client. i tried to install fah6 into another directory and run alongside but it just changed the settings for the original. what am i doing wrong here..

Edited by fizzionz, 10 January 2009 - 04:09 AM.


#25 dnamechanic

  • Topic Starter
  • Life Member
  • 1,518 posts
  • 0
  • Location:Dallas, TX

Posted 10 January 2009 - 01:44 PM

ok so, ive got t he quad running...

Yes, thanks for your contributions to the team.

....but i install the affinity changer but cannot run two instances of the client. i tried to install fah6 into another directory and run alongside but it just changed the settings for the original. what am i doing wrong here...

Your described symptoms:

"...just changed the settings for the original..."

Usually indicates that the OS, in your case Windows, cannot distinguish between the two client installs.

The reason is that the executables (.exe), that is the clients, in both directories have exactly the same name.

To avoid the confusion for the OS, the clients have to be started with a command that tells the OS where to look for the executable (.exe).

This command can be issued as a "flag", in this case the "-local" flag.

To use the "-local" flag the clients must be started with the "-local" present in the start commands.

To do this:

Stop both clients (either with the Graphical control panel or with "Command-C" if using the console (command-line version).

Create a shortcut from each executable to your desktop or wherever you choose.

The executable is located in the client folders and probably looks something like this:

Folding@home-Win32-x86.exe

Right-click each executable and drag to desktop, if that is where you want to have the "Start" shortcuts.

When you release the "right-click", a menu presents, choose "Create Shortcuts Here".

Then "Right-Click" the newly created shortcut and from the menu, select "Properties".

Then in the Target box, add the "-local" flag.

Place one "space" between the flag and the end of the executable (.exe)

Before adding the flag, it could look something like the following example:

C:\FAH6\Folding@home-Win32-x86.exe

After adding the flag, it could look like this:

C:\FAH6\Folding@home-Win32-x86.exe -local

Repeat the above for each client (i.e. two clients, two short-cuts, each to have a -local flag).

After adding several flags (one space between flags) it could look like this:

C:\FAH6\Folding@home-Win32-x86.exe -smp -advmethods -verbosity 9 -local

The other optional flags (as shown above) are:

-smp (only use when requesting SMP work unit, for multi-core only)
-advmethods (sometimes useful to tell Stanford you are willing to work new projects)
-verbosity 9 (tells the client to produce a log file with most detail, i.e. verbose)
-local (tells the OS where to look for an executable file)

A couple of posts that may be helpful for reference:

Brief Description of Contents of folding Directory (Folder) & Multi clients for Mac Pro

Setting up Client for Advanced Methods (Visual aid to flag setup)

#26 fizzionz

  • Guest, F@H
  • 32 posts
  • 0

Posted 11 January 2009 - 02:57 AM

Thank you for the help! ;) all is working perfect.

#27 dnamechanic

  • Topic Starter
  • Life Member
  • 1,518 posts
  • 0
  • Location:Dallas, TX

Posted 18 January 2009 - 03:31 PM

Thank you for the help! :~ all is working perfect.

Fizzionz, thank you for your excellent contributions to the team.




2 user(s) are reading this topic

0 members, 2 guests, 0 anonymous users